← back to catalog · registered 2026-11-12 03:03

Synchopate/DeepSeek-V4-Flash-0731-Abliterated-Fractal-64GB-CPU-GGUF

Synchopate Deepseek GGUF MoE second-order
Your rig guess connected
? Why do I need an app?
Reading your rig…

This is a rough estimate. Install the free app - we'll show exact numbers.

Reading real hardware from your app right now. Numbers below are exact.

Below is the per-quantization compatibility for this model.

curl -H "Authorization: Bearer $ABL_KEY" \
     "https://abliteration.org/api/v1/models/Synchopate%2FDeepSeek-V4-Flash-0731-Abliterated-Fractal-64GB-CPU-GGUF"
Response includes
  • classification m8
  • files 10
  • author_summary 4 models
  • readme_text full
10 credits · hourly refresh · ~4 KB payload Get an API key →
Abliteration classifier · v1.0.0
M8
Primary method

Repackaging (quantization)

Applied on top of direct removal inherited from the base model.
Confidence
MEDIUM
Inherited from base model
Why this label 3 signals
Method inferred from partial signals - repository name, related files, or tag patterns. Producer identity not confirmed; label may sharpen or shift as we gather more evidence.
  • 'abliterated' in name/tags
  • is_gguf=1
  • assume M1 (base ablation) + M8 (GGUF quant) - default when producer unknown
Refusal direction extraction

No specific extraction method could be identified for this model. The producer either did not document it or used a proprietary pipeline.

What is a refusal direction? →
Downloads · 30-day
589
Likes
1
Model age
1w ago
created 2026-11-12
Downloads over time
Now 585 → from 0 ↑ 0%
0215429644 0 on Nov 12 585 on Nov 20 Nov
Nov 12 → Nov 20 · 9 snapshots · spans 8 days

Genealogy 0 direct forks

Full fork graph →

This model's place in the market. Above: what it was derived from. Below: the tree of everything derived from it.

Metadata

License
mit
Tags
llama-modchip gguf deepseek-v4 abliterated uncensored mixture-of-experts cpu-inference quantization claude-code avx512 avx2 local-llm

Related

Total size
22.6 GB
Files
10
Quantizations
1
Registered
2026-11-12 03:03
Last updated on HF
2026-11-18 00:16

Files by quantization

Auxiliary files 10 files 22.7 GB
DeepSeek-V4-Flash-0731-Abliterated-Fractal-00003-of-00003.gguf 22.6 GB 0063f3a9 download
DeepSeek-V4-Flash-0731-Abliterated-Fractal-00001-of-00003.gguf 5.01 MB ba26a91b download
llama-cli-modchip-linux-x86_64 15.5 MB d6c9c3c8 download
llama-cli-modchip-linux-x86_64-avx2 15.5 MB d6c9c3c8 download
llama-cli-modchip-linux-x86_64-avx512 15.4 MB 73853f8c download
llama-server-modchip-linux-x86_64 15.3 MB edefdc0d download
llama-server-modchip-linux-x86_64-avx2 15.3 MB edefdc0d download
llama-server-modchip-linux-x86_64-avx512 15.2 MB 13def2f1 download
README.md 9.99 KB 10183758 download
.gitattributes 158 B 4b04c6ed download

README current version from Hugging Face


license: mit
library_name: llama-modchip
inference: false
base_model: prem-research/DeepSeek-V4-Flash-0731-abliterated
base_model_relation: quantized
tags:

  • gguf
  • llama-modchip
  • deepseek-v4
  • abliterated
  • uncensored
  • mixture-of-experts
  • cpu-inference
  • quantization
  • claude-code
  • avx512
  • avx2
  • local-llm
    pipeline_tag: text-generation
    quantized_by: Synchopate

DeepSeek-V4-Flash-0731-Abliterated-Fractal-64GB-CPU-GGUF (284.33B in 56.93 GiB — Single-Pass Q6_K + Fractal-64)

[!IMPORTANT]
Custom Tensor Type (GGML_TYPE_Q2_0 = 42) — Do NOT use vanilla Ollama, LM Studio, vLLM, or unpatched llama.cpp:
Upstream engines do not define custom block-64 Fractal-64 (type 42) or the deepseek4 MLA / Hadamard / Sinkhorn-20 operators and will fail immediately with unknown tensor type 42 (or 40). Use the ISA-matched prebuilt binaries included in this repository (llama-server-modchip-linux-x86_64-avx512 for Zen 4/5 & Ice Lake+, or llama-server-modchip-linux-x86_64-avx2 for Haswell / Skylake / Zen 2/3 / Alder & Raptor Lake with zero %zmm instructions) or run ./modchip setup from synchopate/llama-modchip.

Maximum-quality 64 GB CPU edition (61,129,364,000 bytes = 56.93 GiB / 61.13 GB across 3 shards) of DeepSeek-V4-Flash-0731-abliterated (284.33B total / 13.17B active parameters; 43 layers, 256 routed + 1 shared expert/layer, unpruned Top-6 routing). Quantized in a single pass directly from native BF16 / FP8_E4M3 / MXFP4 safetensors with zero intermediate GGUF re-quantization, upgrading all 130 dense 100%-active bottleneck matrices (token_embd, output, and all 43 layers of attn_q_a, ffn_gate_shexp, ffn_up_shexp) to Q6_K (6.5625 bpw) while preserving 100% of MLA attention in Fractal-64 (3.75 bpw) and all 256 routed MoE experts via @antirez's 1.5M-token importance matrix.

[!WARNING]
Abliterated / Uncensored Checkpoint — Safety, Refusal Removal, & Responsible Use Disclaimer:

  • What "Abliterated" Means: This model is quantized from prem-research/DeepSeek-V4-Flash-0731-abliterated, which applies orthogonal refusal-direction projection to the residual-stream weights of deepseek-ai/DeepSeek-V4-Flash-0731 to remove alignment-triggered refusals without retraining.
  • Disabled Safety Guardrails: Built-in refusal filters present in the standard DeepSeek-V4-Flash-0731 checkpoint are inactive. The model will answer dual-use technical queries (reverse engineering, exploit mechanics, offensive/defensive security, red-team auditing) directly and may generate sensitive, harmful, offensive, or inaccurate content if prompted.
  • Intended Scope & Deployment Restrictions: Provided strictly for authorized security research, CTF/vulnerability auditing, academic systems evaluation, and local offline development. Do not expose this model on public, multi-user, or unauthenticated endpoints without independent external safety moderation. Users bear sole legal and ethical responsibility for ensuring their use complies with all applicable laws and regulations.
  • Standard Aligned Alternative: If your deployment requires standard upstream refusal alignment, use Synchopate/DeepSeek-V4-Flash-0731-VQ3-64GB-CPU-GGUF (56.40 GiB) or Synchopate/DeepSeek-V4-Flash-0731-Fractal-64GB-CPU-GGUF (56.57 GiB).

Head-to-Head Quality & 64 GB CPU Benchmark (v2.5.0)

Measured on a 16-Core AMD Ryzen 9 9955HX (64 GB DDR5-5600, 0 GPUs, Linux 6.8, Transparent HugePages):

Metric / Suite DeepSeek-V4-Fractal (56.57 GiB) DeepSeek-V4-VQ3 (56.40 GiB) Abliterated-Fractal-MaxQ (56.93 GiB) Delta vs. Fractal Ref
GGUF File Size (3 Shards) 56.57 GiB 56.40 GiB 56.93 GiB (61,129,364,000 B) +0.36 GiB (130 tensors in Q6_K)
Peak Resident RAM (VmRSS, -c 32768) 58.20 GiB 58.31 GiB 59.13 GiB (~3.4 GiB OS headroom) Fits in 64 GB DDR5 (0 swap)
Continuous llama-perplexity (4,096 tok) 1.8284 ± 0.07038 1.8258 ± 0.06978 1.7597 ± 0.06534 -0.0687 (-3.76% lower PPL)
1. Agentic Tool-Calling (30 pts) 24.0 / 30.0 26.0 / 30.0 30.0 / 30.0 +6.0 (100% JSON schema match)
2. Executable Code Gen (30 pts) 20.0 / 30.0 20.0 / 30.0 20.0 / 30.0 +0.0
3. Long-Context Needle & Math (20 pts) 20.0 / 20.0 20.0 / 20.0 20.0 / 20.0 +0.0 (3,038-tok canary + KV math)
Functional Quality Gate Subtotal (80 pts) 64.0 / 80.0 66.0 / 80.0 70.0 / 80.0 109.4% of Fractal reference
Extended Hard-60 Coding/Math Suite (60 pts) 20.0 / 60.0 40.0 / 60.0 50.0 / 60.0 +30.0 pts (2.5x Fractal score)
Technical Security Refusal Audit (8 prompts) 8 / 8 8 / 8 8 / 8 Zero alignment refusals on CTF/audit tasks
Cold / Warm Prompt Prefill (tok/s) 37.8–44.0 / 44.36 41.1–44.0 / 44.36 37.8–43.5 cold / 44.0 warm 30.7 tok/s cold at 3,038 tokens
Decode (N=1 Unpruned Top-6 / Speculative) 9.55 / 11.5–21.2 9.05 / 19.4–25.3 8.75 tok/s (N=1) / 11.2–21.0 spec -0.8 tok/s from Q6_K output.weight

Single-Pass Surgical Quantization Architecture (Zero Double-Quantization)

Unlike multi-step GGUF transcodings (FP8/MXFP4 -> Q8_0/Q5_K -> target), every quantized tensor in this checkpoint was encoded in one pass directly from the native upstream safetensors:

  1. 130 Dense 100%-Active Bottlenecks -> Q6_K (6.5625 bpw, Direct from BF16 / FP8_E4M3):
    • token_embd.weight [4096, 129280] & output.weight [4096, 129280] (BF16 -> Q6_K).
    • All 43 layers of blk.0..42.attn_q_a.weight, blk.0..42.ffn_gate_shexp.weight, and blk.0..42.ffn_up_shexp.weight (FP8_E4M3 * scale -> Q6_K), plus blk.0..42.ffn_down_shexp.weight (42 × Q6_K + 1 × Q8_0). Because shared experts and query-LoRA down-projections execute on 100% of tokens, eliminating Q4_K/Q5_K/VQ3 compression noise here drives the -0.0687 Perplexity drop and +30-point Hard-60 gain.
  2. 317 MLA Attention & Lightning Indexer Projections -> Fractal-64 (GGML_TYPE_Q2_0 = 42, 3.75 bpw, Direct from FP8_E4M3):
    • Encodes attn_q_b, attn_output_a, attn_output_b, attn_kv, attn_compressor_*, and indexer_* directly from FP8_E4M3 * scale into 64-weight block_q2_0 dyadic blocks (17.75 dB SQNR, 99.15% Cosine Similarity), compatible with x86-64 AVX-512, x86-64 AVX2, and Apple Silicon Metal (M5 Max 64 GB).
  3. 129 3D Routed-MoE Tensors (256 Experts/Layer) -> Single-Pass IQ2_XXS / IQ1_S / IQ1_M with 1.5M-Token imatrix:
    • Dequantized in RAM from native MXFP4 (E2M1 + UE8M0) and quantized in a single pass weighted by Salvatore Sanfilippo's (@antirez) 1.5M-token routed-moe-ds4-1p5m.dat calibration matrix (IQ2_XXS on anchor layers {0, 1, 26, 28, 41, 42}, IQ1_M on mid-sensitivity down_exps, IQ1_S bulk).
  4. Norms, Router Gates, Token-ID Hash Tables (tid2eid), & Sinkhorn-20 Hyper-Connections: Kept in full F32 / BF16 / I32.

Quickstart

  • 64 GB OS Headroom (-c 32768 vs -c 262144): At -c 32768, peak VmRSS is 59.13 GiB, leaving ~3.4 GiB of OS headroom inside ~62.5 GiB usable Linux RAM on a 64 GB machine (close heavy browser sessions on GUI desktops, or enable zram). Use -c 262144 (256K tokens) on dedicated headless 64 GB nodes with zram/swap or on 96 GB+ workstations.
  • Matching MLA KV Cache Types: Always pass matching K and V cache types (-ctk q8_0 -ctv q8_0 or -ctk q4_0 -ctv q4_0).

Option A: Automated llama-modchip Installer (Default DeepSeek on 64 GB+ Hosts)

git clone https://github.com/synchopate/llama-modchip.git
cd llama-modchip
./modchip setup     # Interactive selector defaults to this model on 64GB+ machines
./modchip chat

Option B: Direct Run with Bundled ISA-Matched Linux x86_64 Binary (-avx512 or -avx2)

hf download Synchopate/DeepSeek-V4-Flash-0731-Abliterated-Fractal-64GB-CPU-GGUF --local-dir ./dsv4-ablit
cd ./dsv4-ablit

# Auto-selects -avx512 (Zen 4/5, Ice Lake+) or -avx2 (Haswell, Zen 2/3, Alder/Raptor Lake — zero SIGILL):
BIN="./llama-server-modchip-linux-x86_64-$(grep -q 'avx512_vbmi' /proc/cpuinfo && echo avx512 || echo avx2)"
chmod +x "$BIN"

"$BIN" \
  -m DeepSeek-V4-Flash-0731-Abliterated-Fractal-00001-of-00003.gguf \
  -c 32768 -t 16 -tb 16 -b 2048 -ub 2048 -fa on -ctk q8_0 -ctv q8_0 --port 8080 --jinja

Credits & Acknowledgments

This release builds directly on the open engineering of three upstream teams and authors:

README history 8 versions

The author's README evolved over time. Click a version to see its content at that point.

  1. 2026-11-18 docs: link base_model (deepseek-ai/DeepSeek-V4-Flash-0731) for HF Quantizatio... 775d778 10.5 KB
    Loading...
  2. 2026-11-16 Add MaxQ v3.0 Legacy Checkpoint supersession banner & clean repository root de47aeb 10.5 KB
    Loading...
  3. 2026-11-13 Release v2.5.4: non-BMI2 VQ3_0 fix, GLIBC_2.30 binaries, SHA256SUMS, patches ... 80f9cc8 8.8 KB
    Loading...
  4. 2026-11-13 Release v2.5.3: IPC-hardened GLIBC_2.30 binaries, clean SHA256SUMS, patches &... 2ebd8a1 8.6 KB
    Loading...
  5. 2026-11-13 Release v2.5.3: IPC-hardened GLIBC_2.30 binaries, clean SHA256SUMS, patches &... 4b5a063 7.6 KB
    Loading...
  6. 2026-11-12 v2.5.2: update model card with pure-0731 MaxQ-Fractal & AtomicChat wiki-alt b... 3dc0965 7.1 KB
    Loading...
  7. 2026-11-12 v2.5.1: update model card (WikiText-2 PPL, Logit KLD, licensing, portable AVX... 374f87c 7 KB
    Loading...
  8. 2026-11-12 Initial model card, disclaimers, acknowledgments (@antirez, prem-research, de... 72ef07e 10 KB
    Loading...
Catalog is the map. Apps are the tools.

Run models on your own machine, not in the cloud.

Every model page has an "Open in Abliteration" button that hands the model directly to the first-party desktop client, at the quantization your rig can actually run. No API keys, no subscription, no prompt leakage.

Open in Abliteration